Objective: Classical conditioning features prominently in many etiological accounts of panic disorder. According to such accounts, neutral conditioned stimuli present during panic attacks acquire panicogenic properties. Conditioned stimuli triggering panic symptoms are not limited to the original conditioned stimuli but are thought to generalize to stimuli resembling those co-occurring with panic, resulting in the proliferation of panic cues. The authors conducted a laboratory-based assessment of this potential correlate of panic disorder by testing the degree to which panic patients and healthy subjects manifest generalization of conditioned fear. Method: Nineteen patients with a DSMIVTR diagnosis of panic disorder and 19 healthy comparison subjects were recruited for the study. The fear-generalization paradigm consisted of 10 rings of graded size presented on a computer monitor; one extreme size was a conditioned danger cue, the other extreme a conditioned safety cue, and the eight rings of intermediary size created a continuum of similarity from one extreme to the other. Generalization was assessed by conditioned fear potentiating of the startle blink reflex as measured with electromyography (EMG). Results: Panic patients displayed stronger conditioned generalization than comparison subjects, as reflected by startle EMG. Conditioned fear in panic patients generalized to rings with up to three units of dissimilarity to the conditioned danger cue, whereas generalization in comparison subjects was restricted to rings with only one unit of dissimilarity. Conclusions: The findings demonstrate a marked proclivity toward fear overgeneralization in panic disorder and provide a methodology for laboratory-based investigations of this central, yet understudied, conditioning correlate of panic. Given the putative molecular basis of fear conditioning, these results may have implications for novel treatments and prevention in panic disorder.